This is the source code of the middleware layer for the IBM ITK-SNAP distributed segmentation service (DSS) adapted from the open source ITK-SNAP DSS project.
ITK-SNAP is an interactive tool for segmentation of volumetric medical imaging datasets, like CT and MRI. DSS is a web-based application that allows data-scientists to make their advanced image processing algorithms available as services to ITK-SNAP users.
Web Server
The web server as two main role in the Distributed Segmentation Service:
- It's a web interface to visualize and manage available services, providers, ticket history and admin users. It also gives some informations about the different available services and about the DSS itself. The admin access is normally handled with OAuth2 but for testing purpose, a default user
[email protected]have been created. - It's a RestAPI server that can receive tickets from ITK-Snap clients. It handles the different steps of the ticket processing logic and store them in a PosgresDB.
Dispatcher
The dispatcher is a set of deamons (one per service) that regularly check for new tickets on the server. When a ticket is available it is claimed by the corresponding deamon and the deamon handles all the steps required for ticket processing (data download, inference, workspace update, data upload).
Inference server
The inference server is the server that perform the different inference tasks (at this stage: lung-segmantation and nodule-detection). It exposes a REST API that can be reached by the dispacher through port 8080.
This project have been deployed on an Open Shift cluster running with v3.11.161.
We give some example of the commands that we used on out Open Shift Container Platform (OCP) to deploy our project in the following parts, but this documentation aims to be enough generoc so you can adapt the deployment steps to match your own infrastructure.
Each components of the architecture is fully Dockerised and have specific requirements:
- Web Server
- No specific hardware requirement
- Persistent volume
- Expose port
8080to an accessible route
- DB
- No specific hardware requirement
- Dispatcher
- No specific hardware requirement
- Persistent volume
- Inference Server:
- Power node (AC922) with a compatible GPU (see docker image requirements for more informations)
- Persistent volume mounted at
/wmlce/data/output - Expose port
8080to an accessible route
In order to avoid data exchanges between the Dispatcher and the Web Server, we set up a shared storage volume between the two components in order to make both of them access tickets inputs and outputs. This shared volume needs to be created by setting up a persistent volume within the project that you can then mount into the two components.
Details about the directory where this persistent volume needs to be mounted are given in the following parts.
Some of the components exposes ports (8080 for REST APIs). Theses ports needs to be routed into an accessible URL and the corresponding URLs need to be set up in some environment variables (written ENV_VARIABLES) for the different components
- ITK-SNAP Server route: route used to access the web-interface and contact de web server API (ours is
http://itk.10.7.11.23.nip.io)- Needs to be updated in ITK-SNAP DSS service interface
- Needs to be updated in Dispatcher config :
SERVER_URL
- Inference Server route: route used to contact Inference Server API (ours is
http://inference-server.10.7.11.23.nip.io)- Needs to be updated in Dispatcher config:
LUNG_SEGMENTATION_SERVICE_URL: route to inference server (ours ishttp://inference-server.10.7.11.23.nip.io)NODULE_DETECTION_SERVICE_URL: route to inference server (ours ishttp://inference-server.10.7.11.23.nip.io)
- Needs to be updated in Dispatcher config:
- Dispatcher deployment: Create app from Dockerfile
dispatcher/Dockerfile.- OCP example :
oc new-app --name itk-snap-dispatcher https://itk-snap-dss:[email protected]/PSLC/ia-medical/itk-snap-dss.git --context-dir dispatcher - Set-up shared volume: mount persistent volume in
/datastore - Set-up environment variables:
SERVER_URLLUNG_SEGMENTATION_SERVICE_URLNODULE_DETECTION_SERVICE_URL
- OCP example :
- PosgresDB deployment: Create app from Dockerfile
db/Dockerfile(note that db is not persistent in this configuration) .- OCP example:
oc new-app --name itk-snap-dc https://itk-snap-dss:[email protected]/PSLC/ia-medical/itk-snap-dss.git --context-dir db - Set-up environment variables:
POSTGRES_HOST_AUTH_METHOD=true
- OCP example:
- Server deployment: Create app from Dockerfile
web-server/Dockerfile.- OCP example
oc new-app --name itk-snap-web-server https://itk-snap-dss:[email protected]/PSLC/ia-medical/itk-snap-dss.git --context-dir web-server - Set-up shared volume: mount persistent volume in
/app/datastore - Set-up environment variables:
POSTGRES_PORT_5432_TCP_ADDR=itk-snap-dbPOSTGRES_PORT_5432_TCP_PORT=5432ITK_SNAP_SERVER_DATABASE_NAME=admin_dbITK_SNAP_SERVER_DATABASE_USERNAME=admin_userITK_SNAP_SERVER_DATABASE_PASSWORD=(leave empty)ITK_SNAP_SERVER_NOAUTH=1
- OCP example
- Inference Server deployment: Create app from Dockerfile
inference-server/Dockerfile:- OCP example:
oc new-app --name inference-server https://itk-snap-dss:[email protected]/PSLC/ia-medical/itk-snap-dss.git --context-dir inference-server
- OCP example:
In order to register services into to ITK-SNAP server, you need to provide a configuration file and to register a service provider. Configuration files are available in Github repositories to make things easier.
In the admin page of the running DSS server fill the following fields with the relevant information
- New provider :
IBM-LAB - New services assigned to
IBM-LAB:- Provider: newly created provider (IBM-LAB)
- Git repository (Lung segmentation configuration): https://github.com/sinitame/lung-segmentation-service
- Git repository (Nodule detection configuration): https://github.com/sinitame/nodule-detection-service
- Reference: leave blank
If everything is set up properly, you should now see your available services in the Services part of the Web Server available at (ours is http://itk.10.7.11.23.nip.io).
